AI is going to revolutionise procurement. Or so we're told. Every software vendor has added "AI-powered" to their marketing. Every conference features sessions on artificial intelligence transforming supply chains. The hype is considerable.

Beneath the buzzwords, something real is happening. But it's rather less dramatic than the breathless predictions suggest—and rather more useful than the sceptics acknowledge.

What AI Actually Does Well

Strip away the marketing, and AI in procurement does a few things notably well. All of them involve processing large volumes of information faster than humans can.

Document reading is perhaps the most mature application. AI can scan contracts and extract key terms—duration, value, notice periods, liability caps, renewal clauses. What would take a legal team weeks to review manually can be processed in hours.

This isn't magic. The technology is essentially pattern recognition trained on many examples. It makes mistakes, and important contracts still need human review. But for initial triage—identifying which contracts need attention—it's genuinely useful.

Spend categorisation is another strength. Messy procurement data—inconsistent supplier names, unclear descriptions, miscoded purchases—has plagued organisations for decades. AI can learn categorisation rules from examples and apply them consistently across millions of transactions. The result isn't perfect, but it's often better than manual approaches and vastly faster.

Anomaly detection works similarly. What purchases look unusual? Which invoices don't match typical patterns? Which supplier price increases exceed normal parameters? AI can flag outliers for human investigation rather than requiring humans to review everything.

The Practical Use Cases

Beyond these core capabilities, specific procurement applications are emerging.

Contract analysis at scale enables things that weren't practical before. Imagine quickly identifying every contract containing force majeure clauses before a crisis hits. Or finding all agreements that reference a specific supplier being acquired. Or spotting unusual liability terms across your entire contract portfolio.

These searches were theoretically possible before, but practically infeasible. Nobody had time to read every contract looking for specific provisions. AI makes comprehensive analysis possible.

Requirement drafting is an interesting application. Given a set of needs and examples of previous specifications, AI can draft initial requirements documents. Human refinement is essential, but starting from a draft beats starting from blank paper.

Supplier response evaluation shows promise. When tenders generate hundreds of pages of responses, AI can help identify relevant information, flag inconsistencies, and summarise key points. The evaluation judgment remains human, but the preparation becomes faster.

Predictive analytics for supplier risk is increasingly sophisticated. By analysing patterns in financial data, news, and operational indicators, models can flag suppliers showing early signs of distress before problems become obvious.

What AI Doesn't Do Well

Equally important is recognising what AI can't do—at least not yet, and possibly never.

Negotiation requires human judgment, relationship awareness, and contextual understanding that current AI lacks. You can use AI to prepare for negotiations—analysing comparable deals, identifying leverage points—but the negotiation itself remains fundamentally human.

Strategic decision-making involves weighing factors that resist quantification. Should we single-source for cost or dual-source for resilience? Should we pursue this market opportunity or that one? These decisions require judgment that AI cannot provide.

Relationship management is inherently human. Trust, empathy, and partnership don't emerge from algorithms. The supplier who goes extra mile during a crisis does so because of relationship, not contractual obligation.

Novel situations defeat pattern-recognition approaches. AI learns from historical data. When circumstances are genuinely unprecedented—as recent years have repeatedly demonstrated—historical patterns may not apply. Human judgment remains essential precisely when AI is least reliable.

The Implementation Reality

Organisations attempting AI implementations in procurement often discover that the technology is the easy part. The hard parts are more mundane.

Data quality matters enormously. AI trained on bad data produces bad results. Many procurement datasets are incomplete, inconsistent, and poorly structured. Cleaning data for AI consumption often takes longer than implementing the AI itself.

Integration with existing systems is frequently underestimated. AI tools need to connect with ERP systems, contract repositories, supplier databases, and workflow platforms. These integrations are technically challenging and organisationally complex.

Change management determines whether implementations succeed. Tools that procurement professionals don't use deliver no value. Building comfort, demonstrating benefit, and addressing concerns takes sustained effort.

Expectation management is essential. Senior stakeholders expecting magical transformation will be disappointed. Practitioners expecting no value will find reasons to resist. Realistic expectations—meaningful improvement in specific areas, not revolution—support sustainable adoption.

Where to Start

Organisations considering AI in procurement should focus on specific, well-defined problems rather than general "AI transformation."

Identify pain points where volume and repetition are the issue. Contract review taking too long? Spend data perpetually messy? Invoice matching consuming too much effort? These are good candidates for AI assistance.

Start with bounded experiments. A pilot project on contract analysis for one category. A trial of automated spend categorisation for one business unit. Learning what works—and what doesn't—in your specific context beats theoretical planning.

Invest in data foundations. Whatever AI applications you pursue, they'll work better with clean, structured, complete data. Improving data quality is valuable regardless of specific AI plans.

Build internal capability. Even if you buy AI tools from vendors, you need people who understand enough to evaluate, implement, and optimise them. This expertise doesn't appear automatically.

The Honest Assessment

AI in procurement is neither revolutionary transformation nor empty hype. It's a set of tools that do specific things well, particularly involving high-volume document processing and pattern recognition.

Organisations that approach AI pragmatically—identifying specific applications, investing in data quality, managing expectations realistically—will find genuine value. Those expecting AI to transform procurement overnight will be disappointed.

The most valuable immediate applications are unglamorous: reading contracts, categorising spend, flagging anomalies. These boring, high-volume tasks offer immediate ROI and build confidence for more ambitious applications.

Procurement remains a fundamentally human discipline. Relationships, judgment, creativity, negotiation—these core activities aren't being automated. What's changing is the support available for preparation, analysis, and administration. That's less dramatic than the hype suggests, but it's real and it's valuable.